Cluster analysis of comorbidity profiles reveals heterogeneity in hospitalized patients with chronic obstructive pulmonary disease

To the Editor: Chronic obstructive pulmonary disease (COPD) is a global health challenge because of its high prevalence and associated mortality.[1] Multimorbidity is a distinguishing feature of COPD in addition to airflow obstruction. Evaluation from a comorbidity perspective forms part of the comprehensive assessment of COPD. The aim of this study was to analyze the distribution of comorbidities and their impact on the prognosis of patients who are hospitalized with COPD. The study protocol was approved by the ethics board of Peking University Third Hospital (approval number 2022-118-01). The need for informed consent was waived. The study was performed retrospectively and classified patients who were hospitalized with COPD according to their comorbidities. Data held in the electronic medical records in the Inpatient Unit, Department of Respiratory and Critical Care Medicine, Peking University Third Hospital, were reviewed. Patients aged ≥40 years who were admitted between January 1, 2017 and August 1, 2022 with a diagnosis of COPD established by the Global Initiative for Chronic Obstructive Lung Disease (GOLD) criteria were enrolled. Only the index hospitalization was included for subjects with multiple admissions. The following 25 comorbidities recorded in the discharge diagnosis that had a prevalence of greater than 3% were analyzed: hypertension, respiratory infection, arrhythmia, coronary heart disease (CHD), heart failure, diabetes, cerebrovascular disease, benign prostatic hyperplasia, hypoproteinemia, anemia, lung cancer, hyperlipidemia, bronchiectasis, old pulmonary tuberculosis, asthma, atherosclerosis, deep vein thrombosis (DVT), cholelithiasis, interstitial lung disease, gastroesophageal reflux disease, fatty liver, chronic kidney disease, allergic rhinitis, pulmonary embolism (PE), and osteoporosis. Continuous and categorical variables are presented as the median (Q1, Q3) or the number (percentage). Continuous data were compared between the four clusters using the Kruskal–Wallis test and categorical data using the chi-squared test or Fisher’s exact test. All statistical analyses were performed using R version 4.2.0 software (R Foundation for Statistical Computing, Vienna, Austria). Two-sided P-values <0.05 were considered statistically significant. This study included 1127 patients, predominantly ever-smokers and men. The median age was 75 (67, 81) years. During hospitalization, 3.9% (44/1127) of the patients died, and the remaining patients were discharged after receiving treatment. The patient demographics and clinical characteristics are shown in Supplementary Table 1, https://links.lww.com/CM9/C206. Hypertension (52.2%, 588/1127) and respiratory infection (46.9%, 528/1127) were the two most prevalent comorbidities; the prevalence of other comorbidities was less than 30% [Supplementary Table 2, https://links.lww.com/CM9/C206]. Cluster analysis was implemented by principal component analysis and the K-means algorithm. The patients were then divided into four clusters. Table 1 presents the demographic and clinical characteristics of the four clusters. Supplementary Table 2, https://links.lww.com/CM9/C206 shows the prevalence of the comorbidities in the clusters. Cluster 1 (an asthma-COPD overlap [ACO] cluster, n = 100) was characterized by asthma (69.0%, 69/100) and allergic rhinitis (39.0%, 39/100). Cluster 2 (a cluster with less comorbidity, n = 697) had the lowest median age (72 [65, 79] years) and was the largest cluster, accounting for 61.8% (697/1127) of the study population. This cluster also contained the smallest proportion of women (15.4%, 107/697) and had the fewest comorbidities (median 2 [Q1, Q3: 1, 3]). This cluster was not characterized by any specific comorbidity and was used in subsequent analyses as a reference. Cluster 3 (a cardiovascular disease [CVD] cluster, n = 237) contained the oldest subjects (median age 80 years [Q1, Q3: 74, 85]) with CVD, the most common being heart failure (73.4%, 174/237), hypertension (73.0%, 173/237), arrhythmia (60.8%, 144/237), and CHD (57.8%, 137/237). The prevalence of hypoproteinemia (32.1%, 76/237), diabetes (31.6%, 75/237), and anemia (30.8%, 73/237) also exceeded 30%. Cluster 4 (a venous thromboembolism [VTE] cluster, n = 93) included patients with DVT (76.3%, 71/93) and PE (41.9%, 39/93). In cluster 4, apart from DVT and PE, the prevalence of other CVDs, like hypertension, arrhythmia, CHD, and heart failure, fell between the levels observed in clusters 2 and 3. Table 1 - Demographic characteristics and clinical characteristics of the four clusters. Characteristics Cluster 1 (n = 100) Cluster 2 (n = 697) Cluster 3 (n = 237) Cluster 4 (n = 93) Statistics P-value Female 27 (27.0) 107 (15.4) 64 (27.0) 22 (23.7) 20.733* <0.001 Age (years) 74 (65–80) 72 (65–79) 80 (74–85) 78 (72–85) 98.477† <0.001 BMI (kg/m2) 23.8 (20.8–26.6) 22.5 (19.9–25.2) 23.6 (20.2–26.7) 24.1 (20.7–27.0) 15.177† 0.002 Smoking status (ex or current smoking) 78 (78.0) 589 (84.5) 196 (82.7) 73 (78.5) 4.252* 0.236 Number of comorbidities 4 (3–5) 2 (1–3) 6 (4–7) 5 (4–6) 482.360† <0.001 Symptoms Dyspnea 78 (78.0) 532 (76.3) 192 (81.0) 83 (89.2) 9.224* 0.026 Sputum 88 (88.0) 621 (89.1) 209 (88.2) 86 (92.5) 1.416* 0.702 Cough 93 (93.0) 651 (93.4) 215 (90.7) 87 (93.5) 2.001* 0.572 Wheezing 60 (60.0) 248 (35.6) 81 (34.2) 28 (30.1) 26.001* <0.001 Chest tightness 20 (20.0) 88 (12.6) 35 (14.8) 16 (17.2) 4.950* 0.176 Lower extremity edema 7 (7.0) 65 (9.3) 51 (21.5) 18 (19.4) 30.784* <0.001 Blood examination WBC (×109/L) 6.98 (5.42–9.65) 6.96 (5.50–8.96) 7.43 (5.67–9.95) 7.90 (5.72–10.40) 6.812† 0.078 NEU (×109/L) 4.72 (3.67–7.20) 4.76 (3.56–6.82) 5.50 (3.97–7.98) 5.82 (3.86–8.45) 18.718† <0.001 EOS (×109/L) 0.11 (0.02–0.22) 0.12 (0.04–0.21) 0.07 (0.01–0.16) 0.07 (0.01–0.19) 25.820† <0.001 LYM (×109/L) 1.19 (0.91–1.83) 1.27 (0.88–1.70) 1.02 (0.71–1.44) 1.06 (0.73–1.44) 28.968† <0.001 NLR 3.7 (2.3–6.6) 3.6 (2.4–6.8) 5.2 (3.1–9.6) 5.3 (3.1–10.3) 36.984† <0.001 AECOPD 44 (44.0) 328 (47.1) 126 (53.2) 38 (40.9) 5.237* 0.155 Emergency department visit 29 (29.0) 170 (24.4) 116 (48.9) 44 (47.3) 60.391* <0.001 ICU admission 21 (21.0) 145 (20.8) 110 (46.4) 44 (47.3) 76.660* <0.001 Mechanical ventilation 17 (17.0) 120 (17.2) 96 (40.5) 39 (41.9) 71.786* <0.001 Expense (CNY) 15,966 (11,874–22,901) 17,658 (12,227–26,689) 24,888 (15,050–52,899) 27,860 (17,369–55,916) 77.805† <0.001 Death 1 (1.0) 12 (1.7) 23 (9.7) 8 (8.6) <0.001‡ 365-day readmission (n = 918)§ 20 (24.4) 161 (27.1) 56 (31.6) 23 (35.4) 3.614* 0.306 Data are presented as the median (Q1, Q3) for quantitative variables and number (percentage) for discrete variables. *χ2 value. †H value. ‡P-value calculated by Fisher’s exact test. §Only 918 subjects who were admitted before August 1, 2021 and survived hospitalization were included in the readmission analysis. There were 82 patients in cluster 1, 594 in cluster 2, 177 in cluster 3, and 65 in cluster 4. AECOPD: Acute exacerbation of chronic obstructive pulmonary disease; BMI: Body mass index; CNY: Chinese Yuan; EOS: Eosinophils; ICU: Intensive care unit; IQR: Interquartile range; LYM: Lymphocytes; NEU: Neutrophils; NLR: Neutrophil to lymphocyte ratio; WBC: White blood cells. Cluster 1: An asthma-COPD overlap cluster; Cluster 2: A cluster with less comorbidity; Cluster 3: A cardiovascular disease cluster; Cluster 4: A venous thromboembolism cluster. The most common symptoms in all clusters were dyspnea, cough, and sputum. Patients in cluster 1 were more likely to have wheeze than those in cluster 2 (60.0% [60/100] vs. 35.6% [248/697]). There were more patients with lower extremity edema in clusters 3 and 4 than in cluster 2 (21.5% [51/237] and 19.4% [18/93] vs. 9.3% [65/697]). The laboratory results were similar between clusters 1 and 2. The median neutrophil to lymphocyte ratio was higher in clusters 3 and 4 than in cluster 2 (5.2 [Q1, Q3 3.1, 9.6] and 5.3 [Q1, Q3: 3.1, 10.3] vs. 3.6 [Q1, Q3: 2.4, 6.8]), indicating higher levels of systemic inflammation. The incidence of acute exacerbations of chronic obstructive pulmonary disease (AECOPD) ranged from 40.9% (Cluster 4, 38/93) to 53.2% (Cluster 3, 126/237) and was comparable between the clusters. Use of health care resources, including mechanical ventilation, intensive care, and emergency care, was almost twice as high in clusters 3 and 4 than in cluster 2, leading to significantly higher costs (median 24,888 Chinese Yuan [CNY] [Q1, Q3: 15,050, 52,899] and 27,860 CNY [Q1, Q3: 17,369, 55,916] vs. 17,658 CNY [Q1, Q3: 12,227, 26,689]). Mortality was more than five times higher in clusters 3 and 4 than in cluster 2 (9.7% [23/237] and 8.6% [8/93] vs. 1.7% [12/697]). Mortality was lower in cluster 1 than in cluster 2 (1.0% [1/99] vs. 1.7% [12/697]), but the difference was not statistically significant (P = 1.000). Clusters 1 and 2 performed similarly during hospitalization whereas clusters 3 and 4 tended to be worse. For those discharged on medical advice, the 365-day readmission rate ranged from 24.4% (Cluster 1 [20/82]) to 35.4% (Cluster 4 [23/65]) across the clusters but without any significant differences among the clusters. This study classified the COPD population into four clusters with distinct clinical implications. Subjects in cluster 2 had the fewest comorbidities and conditions that were controllable during hospitalization. Cluster 1 consisted of patients with ACO who had clinical outcomes comparable with those of patients in cluster 2. Cluster 3 was characterized by CVD and cluster 4 by VTE. Both clusters presented challenges during hospitalization. It is debatable whether patients with ACO have a worse prognosis than those with COPD alone. The present study suggested that with similar utilization of medical resources, the mortality in the ACO cluster was not significantly inferior to that in the control cluster (1.0% vs. 1.7%). Other researchers have reported similar findings of lower mortality and a better prognosis in patients with ACO during long-term follow-up.[2] We hypothesized that airway obstruction would be more reversible and more responsive to treatment in patients hospitalized for respiratory symptoms of ACO than in those hospitalized for COPD alone. Comorbid asthma did not result in higher mortality during hospitalization for COPD. In our study, the CVD and VTE clusters were notable for their marked decline in health status and a mortality rate that was five times that of the control cluster. The cardiopulmonary continuum illustrates the close association between COPD and CVD. Exacerbation of COPD and adverse cardiovascular events are mutual risk factors. In patients with COPD, CVD-related hospitalization and mortality rates are greater than that attributable to COPD itself. Inflammation spilling over from the lung into the circulation was considered to be the key factor responsible for the close relationship between COPD and CVD.[3] Observations of elevated neutrophils and an increased neutrophil-to-lymphocyte ratio in the CVD and VTE clusters were consistent with the process of inflammatory spillage. Hypoxemia-induced pulmonary vasoconstriction and the compressive effects of pulmonary hyperinflation also contribute to worsening cardiovascular status in patients with COPD. Readmission rates have increasingly been recognized as an important indicator of the quality of health care, and some countries are now penalizing hospitals with excessive readmissions after AECOPD. More comorbidities will likely lead to more adverse events and more admissions, whether caused by AECOPD or the comorbidities themselves. In this study, comorbidities significantly affected health care utilization and in-hospital mortality but were not associated with readmission rates, which seems somewhat counterintuitive. Many studies have investigated the effect of comorbidities on the likelihood of readmission in the COPD population. Previous findings have suggested that the effect of comorbidities varies by type; for example, heart failure and renal failure were identified to be risk factors of readmission.[4] In our study, this was further complicated by the fact that we did not study a single comorbidity but a combination of comorbidities. Overall, this study suggested that comorbidities do not affect readmission. As discussed above, comorbidity offers a valuable insight into the heterogeneity of COPD. Some studies have reported that spirometry is not a prognostic factor in patients who are hospitalized for COPD. In the present study, evaluation from a comorbidity perspective had unique advantages. Selection of the included comorbidities did not rely on a priori knowledge, allowed us to explore underassessed comorbidities, and identified novel phenotypes of COPD. VTE was distinguished from CVD and identified as an independent cause of hospitalization. Our previous study investigated phenotypes of COPD in the general Chinese population and emphasized the need for timely detection and appropriate management of COPD in primary care.[5] Our present findings complement our knowledge of COPD phenotypes in a tertiary hospital. This study has several limitations. First, there were no data on admissions to other hospitals, so the readmission rate was probably underestimated. However, our readmission rate is similar to that in previous reports. Second, the prevalence rates of atherosclerosis and osteoporosis were lower than those previously reported. Underdiagnosis of these diseases is common in China and inevitable in retrospective studies. However, the scale-free architecture of the COPD comorbidity network makes comorbidity clustering a robust method even with limited data.[6] Funding This work was supported by the Beijing Nova Program (No. 20220484157), National Natural Science Foundation of China (No. 82090014), Proof of Concept Program of Zhongguancun Science City and Peking University Third Hospital (No. HDCXZHKC2021206), and Clinical Cohort Construction Program of Peking University Third Hospital (No. BYSYDL2021013). Conflicts of interest None.